Research On Recommendation Algorithm Based On Knowledge Graph

With the rapid development of the Internet,people are facing the problem of multiple choices and information overload.The collaborative filtering recommendation algorithm is an effective method to solve this problem,but this method has the problems of sparse interactive data and cold start.To solve this problem,some paper introduces the knowledge graph into the recommendation algorithm to improve the performance of the recommendation algorithm.The KGCN model that introduces the knowledge graph into the recommendation completes the recommendation through aggregating the neighbor information on the product side.However,this model only aggregates the neighbor information on the product side and cannot more effectively explore the potential interests of users.In order to solve this problem,this paper proposes a method of fusing user neighbor information on the user side.First,we obtain the first-order neighbor node and the high-order neighbor node set according to the position of the user’s interaction record in the knowledge graph network.In the relational space of knowledge graph,the attention mechanism is used to spread the user’s interest.After spreading to a certain order of neighbors,a user vector is generated,and finally the user vectors of each order are merged into the final user vector.On the item side,the user vector fused with the user’s neighbor information is sent to the KGCN model to generate the item vector,so as to realize the indirect interaction between the user and the item’s neighbor information,and finally calculate the interaction probability between the user and the item using the inner product operation.Comparative experiments on public data sets have verified the effectiveness of the algorithm.In order to solve the problem that the path-based RKGE model only explores the relationship between users and items based on the order of the sequence,but cannot effectively improve the expressive ability of the model,this paper proposes a recommendation algorithm based on the knowledge graph that merges path and neighbor information.For the acquisition of path vectors,using the idea of the RKGE model,first automatically construct a path based on user interaction records and knowledge graphs,then use LSTM to extract the information of each path,generate a vector for each path,and finally use a pooling operation to generate the final path vector.For obtaining the neighbor vector,first obtain the first-order neighbor set directly connected to the user and the first-order neighbor set directly connected to the item according to the path set of the user and the item obtained in the process of obtaining the path vector.Then obtain high-order neighbors based on these first-order neighbors in the knowledge graph,Through GCN or GAT aggregating neighbor information,the neighbor vector on the user side and the neighbor vector on the item side are obtained,and the two vectors are merged to obtain the neighbor vector.Finally,the neighbor vector and the path vector are spliced into a vector and sent to the prediction link to calculate the interaction probability between the user and the item.Comparative experiments on multiple public data sets have verified the effectiveness of the algorithm.